[4I3-GS-7] Language media processing:

[4I3-GS-7-02] Eye blink detection using personalized eye aspect ratio based on 3D landmark

〇Koki Saito¹ (1. BIPROGY Inc.)

Keywords:blink, facial landmark, eye aspect ratio

Blink detection technology plays an important role in communication and various situations. However, it is not easy to accurately detect blinks because people have different eye size, blink frequency, and blink duration. This paper proposes a blink detection method that takes into account individual differences based on video clip. In this method, eye aspect ratio (an index of eye opening) is calculated by using 3d landmark from video clip, and a personalized eye aspect ratio threshold is calculated by optimizing it for each individual. We evaluated this method on public datasets, and results showed that accuracy, precision, and F1-score were improved compared to previous study. It is found that robust eye aspect ratio to facial motion can be calculated by using 3D landmark and the personalized eye aspect ratio can solve the problem of individual differences.